THE AGE OF INTELLIGENT MACHINES | Chapter 2: Philosophical Roots
September 24, 2001
- Ray Kurzweil
- The MIT Press (1992)
“Some philosophers hold that philosophy is what you do to a problem until it’s clear enough to solve it by doing science. Others hold that if a philosophical problem succumbs to empirical methods, that shows it wasn’t really philosophical to begin with.” — Jerry A. Fodor, Representations: Philosophical Essays on the Foundations of Cognitive Science
“The chance of the quantum theoretician is not the ethical freedom of the Augustinian.” — Norbert Wiener, Cybernetics
“There’s something queer about describing consciousness: whatever people mean to say, they just can’t seem to make it clear. It’s not like feeling confused or ignorant. Instead, we feel we know what’s going on but can’t describe it properly. How could anything seem so close, yet always keep beyond our reach?” — Marvin Minsky, The Society of Mind
How can mind arise from nonmind? In examining human thought through the ages, philosophers appear to have gone down one of two paths. One school of thought, which we might call mind as machine, starts with the observation that human thought takes place in the human brain. The brain, in turn, is made up of tens to hundreds of billions of neurons. Neurons, while not simple structures, can nonetheless be fully understood as biochemical machines. Our brain thus consists of billions of biochemical machines interacting with each other, a fact from which we can draw two conclusions. The first is that the human mind is a machine, albeit an enormously complicated one. It is, after all, made up of matter just as subject to natural laws as any other machine. Second, we can, at least in theory, create other, human-made machines that employ the same techniques or algorithms for transforming information as the human brain. It is thus possible to replicate in a machine intellectual capacities that previously could only be achieved by human intelligence. This latter conclusion is a clear tenet of the AI movement.1
The opposing school of thought, which we can call mind beyond machine, contends that there are certain aspects of human thought and human existence that cannot be understood through this type of analysis. While acknowledging that the model of the human mind as machine may provide some understanding, it cites such attributes as consciousness and free will and, depending on the philosopher, other possible attributes as well, as being outside the fully rational, i.e., machinelike, model. It criticizes the approach of applying only logic to our understanding of thinking as being hopelessly circular, that is, as using an analysis based only on logic to conclude that thinking is based only on logic.2
In this debate the concept of emotions and feelings generally lie in a middle ground somewhere between consciousness and free will on the one hand and the logical patterns of rational thought on the other. The mind-as-machine school tends to analyze emotion as another form of logical thought, subject to its own rules. In this view, our emotions are a complex set of algorithms that motivate our behavior in a way that supports the greater goals of our culture. One could say that our emotions provide us with our strategies, while our more detailed logical calculations provide us with our tactics. The mind-beyond-machine school is more likely to conceive of emotion as being deeply imbedded in our consciousness and thus not fully comprehensible by logic alone.3
Thousands of years before there were computers, Plato (427-347 B.C.) recognized the similarity between at least certain aspects of human thinking and the apparently determined cause-and-effect behavior exhibited by machines.4 The Platonic debates illuminate as clearly as any modern philosopher the apparent paradox of the human mind displaying free will while being subject to predictable natural laws. In the final analysis, it appears that Plato accepted both views as an irreducible duality. Such paradoxes, according to Plato, were not to be avoided; rather, they were the key to a richer understanding of the human condition.
Philosophical schools based on Plato’s thought continued to shape European epistemology-the study of the limits and validity of knowledge-into the middle ages through the lasting influence of the Academy he founded and through one of his star pupils, Aristotle (384-322 B.C.).5
The European Renaissance of the seventeenth century and the Enlightenment, a philosophical movement of the seventeenth and eighteenth centuries, renewed and intensified an emphasis on the scientific method and in particular its application to thinking as a process that could be studied and understood as a phenomenon following natural laws. The rationalism of Rene Descartes (1596-1650), the empiricism of Francis Bacon (1561-1626) and the physics of Isaac Newton (1642-1727) were fused by Immanuel Kant (1724-1804) into a rigorous view of man’s world based on knowledge as its principle building block.6
Modern philosophy has divided into two schools of thought. While often seen as contradicting one another and while frequently clashing with one another indeed, they primarily deal with divergent issues. Existentialism, which is the dominant school of philosophy today in Europe, has its roots in the unorthodox Christianity of Soren Kierkegaard (1813-1855) and the anti-Christianity of Friedrich Nietzsche (1844-1900). Existentialism, and a related field, phenomenology, regard human thought and human existence as describable by scientific analysis, but only partially.7 Their emphasis has been on such phenomena as guilt, anxiety, and suffering, which in their views lie beyond fully rational exploration and are keys to understanding the limits of reason. The modern existentialism of such philosophers and playwrights as Jean-Paul Sartre (1905-1980) and Samuel Beckett emphasize the role of free will in an apparently purposeless world.
A movement still popular in the United States that often clashes with existentialism is logical positivism, based on the early work of Ludwig Wittgenstein (1889-1951) and developed by Alfred Ayer and others. Searching for truth in the foundations of language, logical positivism gave rise to the development of linguistic theory by Noam Chomsky and others and greatly influenced the emergence of computation theory.8
The AI movement, which can be said to have its roots in logical positivism, has often clashed with the phenomenology of the existentialists, as most recently exemplified by the ongoing debate of AI practitioners such as Marvin Minsky, Seymour Papert, and Ed Feigenbaum with leading AI critic and phenomenologist Hubert Dreyfus. Unfortunately, this debate has been overly personal and divisive and thus has not contributed to a needed synthesis between the two pillars of modern philosophy.9
Plato and the Platonists
“The safest general characterization of the European philosophical tradition is that it consists of a series of footnotes to Plato.”
- Alfred N. Whitehead
“All virtue is one thing, knowledge.”
Born in 427 B.C., Plato is regarded by many as the greatest philosopher of all time. His thought ranged across the ultimate nature of truth and knowledge, ethics, social order, political structure, physics, astronomy, and mathematics. His writings not only recorded his thoughts but constitute the primary written record of the teachings of his mentor, Socrates (c. 469-399 B.C.).10 Aristotle viewed his own work as carrying out the dissemination and further development of Plato’s teachings.11 Socrates, Plato, and Aristotle are credited with having established the essentially rationalistic philosophy of Western culture.
While he is now considered one of the greatest writers of philosophy, Plato considered his writings as merely a tool to assist him in his teaching. He regarded his principle work as the establishment and guidance of the Academy, an institute for the pursuit of science and philosophy that he founded at the age of 41. The Academy outlived Plato, and the influence of its students and followers enabled Platonic thought to exert profound influence for many centuries after his death at age 80.12
While Plato had his mystical side, his is a highly ordered mysticism based on a theory of ideal forms, such as the form of the circle, of beauty, of love, etc., and the manifestation of forms in an imperfect world.13 Plato’s own logical reasoning exemplified the power of reason. By logical inference and his own thought experiments, Plato was able to determine, for example, that the planets followed orbits of single closed curves and that other apparent movements of stars and planets were due to the earth’s movement through the sky in its own closed curve.14 He then imagined the earth and the other planets circling a “Pythagorean central fire.”15
Plato inferred the existence of irrational numbers, numbers that could not be expressed as the finite sum of fractions. The square root of 2 is the quintessential example. Plato saw the fact that rational numbers, with their finite definitions, and irrational numbers, with their infinite definitions, coexist in the same continuum as symbolic of the coexistence of material and mystical phenomena in nature.16
Though Plato’s lasting contribution is in the rationalization of philosophy and a casting aside of the ornate mysticism of many of his peers, he maintains that there exists a level of creation that defies complete rational understanding. In his Timaeus, he describes the ananke, the level of reality that cannot be rationalized completely and that has to be accepted as a reflection of the purpose of creation.17 Modern existentialism echoes Plato in its acceptance of a rational level of reality combined with its emphasis on the limits of reason and logic.
At the core of the duality of existence in the rational and mystical is the issue of consciousness and free will. In the Phaedo and later works, including The Republic and Theaetetus, Plato expresses the profound paradox inherent in the concept of consciousness and man’s ability to freely choose. On the one hand, human beings partake of the natural world and are subject to its laws. Our minds are natural phenomena and thus must follow the cause and effect laws manifest in machines and other lifeless creations of man. On the other hand, cause and effect mechanics, no matter how complex, do not, according to Plato, give rise to self-awareness or consciousness.18 Plato attempted to resolve this conflict in his theory of the Forms.19 Consciousness is not an attribute of the mechanics of thinking, but rather the ultimate reality of human existence. Our consciousness or “soul” is immutable and unchangeable. Thus, our interaction with the physical world is on the level of the “mechanics” of our complex thinking process in a complex environment. Yet Plato was not fully satisfied with this metaphysical doctrine. If the soul is immutable and unchanging, then it cannot learn or partake in reason, because it would need to change to absorb and respond to experience. He expressed dissatisfaction with positing consciousness in either the rational processes of the natural world or the mystical level of the ideal Form of the self or soul.
An even deeper paradox is apparent in the concept of free will. Free will is purposeful behavior and decision making. Plato believed in a “corpuscular physics” based on fixed and determined rules of cause and effect.20 If human decision making is based on such interactions of basic particles, our decisions too must be predetermined. Such predetermination would, however, appear to contradict human freedom to choose. The addition of randomness into natural laws is a possibility, but it does not solve the problem. Randomness would eliminate the predetermination of decisions and actions, but it contradicts the purposefulness of free will, as there is nothing purposeful in randomness. Positing free will in the soul and thus separating it from the rational cause and effect mechanics of the natural world is also not satisfactory for Plato, because Plato is uncomfortable placing reason and learning in the soul. These attributes of our thinking process are too orderly and logical to place entirely on the mystical plane.21
From Aristotle on, philosophers have debated for over two thousand years exactly what Plato’s position was on these issues. Support for apparently contradictory views can be found in Plato’s writings. My own view is that Plato believed in an essentially irreducible paradox at the core of the issues of consciousness and free will. I base this on his refusal to write his own metaphysics and on the eloquence with which Plato is able to articulate the alternate sides of each paradox. Plato’s choice of the dialog form was an excellent medium for expressing paradox in that it freed him to express passionately conflicting views.22
Irreducible paradox at the core of reality has found support in a surprising place-twentieth century physics. Physics, which seeks to describe ultimate reality in rational terms, has concluded that the essence of electromagnetic radiation is both a particle and a wave, two mutually exclusive and inconsistent models. Quantum theory too is based on a paradox: particles have precise locations, but we cannot know what these locations are. In its ultimate form, quantum mechanics describes a particle as having no precise location, although somehow the particle exists in a multidimensional space .23
In perhaps his most direct appeal to paradox as a resolution of apparent conflict of ideas, Plato discusses the duality of human love in Phaedrus.24 By applying the logical method to the study of passion, he concludes that love and its expression in the apparently mad behavior of the lover is rooted both in the material world and in the attempt of the soul to achieve union with the ideal Form of transcendent emotion.
What is truly remarkable about Plato’s writings is the extent to which they are, after twenty-three hundred years, relevant to modern philosophical dilemmas on the relationship of human thought to the mechanics of the machine. Plato saw clearly the relationship of human thought to the rational processes of a machine. He recognized that human thought was governed by natural law and that natural law was an essentially logical process. There is no limit, according to Plato, to the extent to which we can unravel human thought and behavior by scientific observation and logical inference. At the same time, he felt that human reality was not sufficiently expressed in logic alone. He does not resolve this problem, however, by attributing human thought to mystical processes that are of a different world from the logical processes of the material world. Instead, he resolves that the duality of human thought as both a logical process and one that transcends pure logic represents a necessary coexistence. It is a synthesis of views that is relevant to the modern conflict between the logical positivist foundation of the AI movement and existential-phenomenological views.
While Plato’s thought had wide and diverse influences and interpretations after his death, one of the more interesting refinements of his thought made by his successors was a certain “mathematization” of his philosophy and the expression of his philosophy of Forms in numeric terms.25 The Forms, which constitute ideals such as round, beauty, justice, and love, are regarded as pure concepts in the same way that numbers are pure concepts. At the other extreme of reality is the physical world that imperfectly manifests the Forms. In between the Forms and physical reality are “mathematicals,” which, like the Forms, are perfect and unchanging but, like physical reality, are numerous. In other words, there is only one concept or Form circle, but there can be many instances of circles. Each instance can be immutable and perfect in the mathematical realm, while each manifestation of circularity in the physical world will be changeable and somewhat less than perfect.
These views are not directly expressed in the dialogues but are often attributed (e.g., by Aristotle) to Plato’s later oral teachings. They take place during a time of fertile development of mathematical theory by members and associates of Plato’s Academy, including Euclid (330-260 B.C.), the expositor of plane geometry, and Theaetetus (c. 415-369 B.C.), the creator of solid geometry.26
The expression of the mystical concept of Forms in the logical language of mathematics expresses again the paradox at the heart of Plato’s views of human reality and thought. In the Epinomis, Plato states that the “relations of numbers are the key to the whole mystery of nature.”27
The Enlightenment, along with parallel fertile developments in science and theology, was a philosophical movement to restore the supremacy of human reason, knowledge, and freedom. It had its roots in both the artistic, literary, and cultural activity of the previous three centuries (the European Renaissance) and the Greek philosophy of twenty centuries earlier. It considered its own roots to be with Socrates, Plato, and Aristotle, and it constituted the first systematic reconsideration of the nature of human thought and knowledge since the Platonists.28
With Isaac Newton’s presentation of the laws of gravitation in his Philosophiae Naturalis Principia Mathematica, published in 1687, as well as advances in the construction of clocks and mechanical automata, the philosophers of the Enlightenment had more powerful models both of natural laws and of the potential of machines than did their counterparts two thousand years earlier.29 Machines became more elaborate and more capable as mechanical automata developed in sophistication and popularity during the seventeenth and eighteenth centuries.30
The Enlightenment saw a blending of philosophy and science, as the same persons often dominated both fields. Descartes, for example, formulated the theory of optical refraction and developed the principles of modern analytic geometry. In his own view, Descartes’s efforts in mathematics and science were intended primarily as a means of exploring and demonstrating certain aspects of his metaphysical doctrine. He needed to demonstrate the deterministic nature of the real world and these major scientific discoveries were in a sense footnotes to Descartes’s philosophical investigations. The mystery of how mind can arise from nonmind, of how thoughts and feelings can arise from the ordinary matter of the brain, sometimes called the mind-body problem, was perhaps most clearly articulated by Descartes.31
In his comprehensive Discourse de la Méthode, Descartes pushed rational skepticism to its limits. Acknowledging that the existence of other people and even our own bodies may be illusions, he concluded that we cannot doubt the existence of our own thought and hence his famous conclusion “I think, therefore I am.”32
Descartes was fascinated by automata and made contributions to their design. Once while Descartes was traveling by sea, the ship’s captain was startled by the realistic movements of Descartes’s mechanical doll Francine and forcibly threw “her” overboard, believing the automaton to have been a product of the devil.33
Sir Isaac Newton
Linking the process of thought to the determined interactions of the natural world gained momentum with Newton’s breakthrough in the understanding of mechanical law. Newton set out as his goal to find a link between the mechanical interactions of objects we observe in the laboratory and the movement of celestial bodies that we observe in the sky. The goal was a unified set of formulas that explains the movement of objects from the very small to the very large, something never before achieved. The result was unexpectedly successful-Newtonian mechanics appeared to explain with extreme accuracy a deterministic order that governed all matter.34 With one publication Newton swept aside centuries of medieval imprecision.
Newtonian mechanics remained the dominant view of both celestial and particle mechanics for almost two and one half centuries.35 While Einstein (1879-1955) showed Newtonian mechanics to be a special case of his broader theory of relativity, this “special case” happens to be the world we live in.36 One has to enter the world of high-speed subatomic particles or certain astronomical phenomena to witness significant deviations from Newtonian mechanics. Studies of naive physics, which is an exploration of the mental models of the physical world that people actually use, indicate that we believe that we live in a Newtonian, not an Einsteinian, world.
When Newton had successfully explained major aspects of the universe, the implications of his theory were not lost on the theologians of his day.37 The success and apparent power of Newton’s ideas were seen as a threat to the unique status of man. Fear and anger were expressed that the same mathematical reasoning that resulted in Newtonian Mechanics would be extended to the processes of the mind. It was feared that human thought would be subjected to the indignity of comparison with the determined interactions of billiard balls, which were often used to demonstrate mechanical principles. Even Wilhelm Leibniz (1646-1716), who shares credit with Newton for the invention of the calculus used to express Newtonian laws, attacked Newton for regarding God as a “clumsy watchmaker.”38
Immanuel Kant, born in 1724 in East Prussia (now part of the Soviet Union), is widely regarded as the preeminent philosopher of the Enlightenment. He typified the Enlightenment in his emphasis on human reason and rationality, and he attempted to develop a metaphysical doctrine based entirely on reason.39 In Kant’s conception, human thought is guided by a priori principles and concepts, that is, concepts and structures that are not based on experience. Furthermore, in Kant’s view, human knowledge constitutes the ultimate reality. This reversed the prior conception that ultimate reality resided in the physical world, with our thoughts based on our imprecise sensory impressions reflecting imperfect models of the physical world.40
Three of Kant’s innovations would profoundly alter the philosophical landscape and set the stage for the emergence of twentieth-century rationalism. First, Kant’s model of a priori concepts influenced and is echoed in the logical positivist search for truth in language and the concept of the innate structures in language postulated by modern linguistics. Second, by rejecting Descartes’s dichotomy between the instinctive reflex of the animal and the rational thought of man, Kant opened up even further than his predecessors the processes of human thought to analytic investigation. Finally, Kant’s emphasis on the supremacy of knowledge over other levels of reality would be echoed by both the logical positivist and existentialist schools of the twentieth century.
The relationship of the Enlightenment to the modern schools
On the issue of the ultimate nature of human reason, the philosophers of the Enlightenment, while diverse in their views, essentially agree with the modern existential school. Both maintain that while logic and the scientific method can be used to explore human thought and the thinking process in great detail, there is a level of human existence and experience that defies such analysis. However, the Enlightenment ended up defining the metaphysical level with such logical rigor, stripping away the complexities of earlier mysticism, that it set in motion a movement to explain all of reality in rational terms. The logical-positivist school continued and extended this tradition.41 The existential movement, on the other hand, can be said to be a reaction to this trend. While acknowledging the role of logic and scientific analysis, the existentialists have attempted to refocus attention on the irrational and paradoxical side of the human condition.
The Logical Positivists and the Existential Reaction
“The myth that everything in the world can be rationally explained has been gaining ground since the time of Descartes. An inversion was necessary to restore the balance. The realization that reason and anti-reason, sense and non-sense, design and chance, consciousness and unconsciousness belong together as necessary parts of a whole.”
-Hans Richter, Dada
“It is essential to abandon the over-valuation of the property of being conscious before it becomes possible to form any correct view of the origin of what is mental.”
-Sigmund Freud, The Interpretations of Dreams
Kant as the first logical positivist
The Enlightenment as it had culminated in the thought of Immanuel Kant had relegated metaphysics to an elegant and simple role in explaining the human condition. Our understanding of human memory, our reasoning faculty, even our emotions could be understood in terms of scientific inquiry and analysis. While Kant did regard the metaphysical level as supreme, he had much more to say about the intricacies of the physical level. Although it was not his intention, his impact has been to move the western philosophical tradition toward less emphasis on the role of metaphysics.”42 He has, in fact, been called the “first logical positivist.”
The title of the first logical positivist truly belongs, however, to Ludwig Wittgenstein. An enigmatic figure, Wittgenstein gave away his large inherited fortune so as not to be distracted from his philosophy by worldly possessions.43 His most influential and first major work, the Tractatus Logico-Philosophicus, was not an instant success.44 Wittgenstein had a great deal of difficulty in finding a publisher for his work, and it was ultimately the influence of his former instructor, Bertrand Russell (1872-1970), that allowed the book to come into print.45 It has come to be regarded by some, however, as perhaps the most influential philosophical work of the twentieth century.
Wittgenstein further applied the analytic treatment of human thought, extended by the Enlightenment, to the study of human language. He examines the nature of language, how it is that we communicate using language, and what it is that we are communicating. His examination is not, however, an exploration of the structure, organization, physiology, or psychology of communication. Rather, it is an attempt to provide a philosophical definition of knowledge-what we can know-by analyzing the meaning of language. He goes on in the Tractatus to consider language to be the embodiment of what can be said, what can be known, indeed, what can be thought:
18.104.22.168.1 All philosophy is a “critique of language.”46
5.6 The limits of my language mean the limits of my world.47
5.6.1 We cannot think what we cannot say.48
Wittgenstein goes on to define language in a particular, some would say peculiar, way. In Wittgenstein’s world there are certain elementary facts, there are propositions about relations between elementary facts, and there are certain allowable transformations on such propositions that yield composite propositions. His concept of human thought is that we receive sense impressions, which comprise elementary facts. We can then transform these elementary facts and derive relationships among them according to the allowable logical processes. Any thought outside this scheme is either false or nonsensical.
Wittgenstein makes two major points that have a direct bearing on the intellectual roots of artificial intelligence. He makes a direct link between human thought and a formal process that can be described only as computation. To reorder Wittgenstein’s statements, we cannot think what we cannot say; we cannot say, or at least we ought not say, what is meaningless in the language we are speaking; statements in any language are indeed meaningless unless they can be derived from a formal (and therefore computation like) sequence of transformations on a data base of elementary propositions.
This description of human thought as a formal sequence of computations would be restated two decades later in the Church-Turing thesis (see chapter 3).49 The Church-Turing thesis was regarded as a radical doctrine when it was first proposed, but it has its roots in the Tractatus. It is not a thesis that everyone familiar with it necessarily accepts, and it remains controversial today. Wittgenstein himself ended up rejecting it.
The other point made in the Tractatus that would have significance later to computational theorists is that thought is embedded in language.50 It is also interesting to note that language as conceived in the Tractatus has more of the quality of the programming language LISP or even PROLOG than it does of Wittgenstein’s native German.
The organization of the book is also interesting. The treatise contains only seven primary formal statements, numbered 1 through 7. To help us along, Wittgenstein also includes several levels of modifying statements. For example, statements 1.1 and 1.2 modify statement 1. In turn 1.1.1 is provided to help explain 1.1, and so on. The reader has the choice of reading the book from left to right or from top down. The modular structure would please today’s proponents of good programming style.
Wittgenstein starts with, “The world is all that is the case,” (statement 1) and ends with, “What we cannot speak about we must pass over in silence” (statement 7).51 One thing that is clear from these two statements is that the Tractatus was an ambitious work.
In 1953, two years after Wittgenstein’s death, Philosophical Investigations, his last work, was published.52 Views of its significance vary with the point of view of the critic. Logical positivists who trace their intellectual roots to the early Wittgenstein regard his later work as confused and disorganized, while existentialists regard it as a work of major importance.53 In what is perhaps a unique occurrence in philosophical history, Wittgenstein is credited with having established two major systems of philosophy, each with great influence, with the second criticizing and rejecting the first.54 Wittgenstein ends up near the end of his life having a lot to say about subjects that he had argued in the early Tractatus should be passed over in silence.
Ayer and the formalization of logical positivismAlfred Ayer and others carried on Wittgenstein’s work, correcting errors and expressing it with a degree of formalism and rigor that Wittgenstein had neither the time nor patience to carry out.55 At a philosophical level, Logical Positivism argues that every statement, and by implication all knowledge, is either based on sense data (“a posteriori” or “synthetic”), or is based on logic (“a priori” or “analytic”). It rejects all metaphysical theories as strictly meaningless and having only emotive force. On a practical level, modern theories of linguistics and computation were derived from the formalisms of Logical Positivism. One does not need to accept all of its implications to recognize its contribution as the philosophical basis of artificial intelligence.
The existential reaction
As with any major philosophical movement, the views and theories of its proponents are diverse and complex. It is nonetheless reasonable to view existentialism as a reaction on an intellectual and cultural level to the major drift of Western thought toward greater and greater reliance on rational and analytic views. At its core, existentialism defines human reality as almost the reverse of the logical-positivist view. It considers the analytic and synthetic types of statements, which are the only meaningful statements in logical positivism, as either meaningless or trivial. It regards the spiritual and emotive life, which are meaningless in logical-positivist terms, as the seat of true meaning.56
Some recent views of the early Wittgenstein claim that he was not denying the existence of a realm beyond the narrow definition of meaning expressed in the Tractatus. Clearly, the last sentence in the Tractatus, “What we cannot speak about we must pass over in silence,” is referring to something.57 If it is referring to something that does not exist, then even by the early Wittgenstein’s own standards the sentence would be meaningless. Since we can assume that Wittgenstein would not end his book with a meaningless sentence, the phrase “what we cannot speak about” is referring to something meaningful. If “what we cannot speak about” is a meaningful concept and yet we must pass over it in silence, then the sentence must be a plea for silent contemplation of a higher realm.
The Debate Goes On
“Why is philosophy so complicated? It ought to be entirely simple. Philosophy unties the knots in our thinking that we have, in a senseless way, put there. To do this it must make movements that are just as complicated as these knots. Although the result of philosophy is simple, its method cannot be if it is to succeed. The complexity of philosophy is not a complexity of its subject matter, but of our knotted understanding.”
- Ludwig Wittgenstein
The relationship of human thought to the “logical” process of the computer continues to be controversial, a continuation of the debate started in the Platonic dialogues. The very name “artificial intelligence” juxtaposes two concepts that engender diverse and often intense intellectual and emotional reactions.
One particularly noisy debate has been going on for twenty years between a number of the academic AI leaders and Hubert Dreyfus, a modern phenomenologist and Berkeley professor. It began with a paper Dreyfus wrote as a consultant to the Rand Corporation in 1965 entitled, “Alchemy and Artificial Intelligence.”58 As might be clear from the title, it was a no-holds-barred attack on what was at that time an uncertain new academic discipline. Since then Dreyfus has made something of a career out of attacking artificial intelligence.
Such criticism might be useful to the field, and perhaps some of it is. Unfortunately, the debate has been marred in a number of ways. First, there has been considerable personal anger expressed on both sides. Dreyfus has been quoted as saying, “Why do I get so upset with people like Papert, Minsky, Newell and Simon?-and I really do get upset. It’s really puzzling. I’ll have to think about that …. Maybe I attack in them what I dislike in myself, an excessive rationality.”59
More serious is an unwillingness on both sides to fully understand the disciplines and traditions of the other. Dreyfus has displayed considerable ignorance of AI methods and status. In a recent article Dreyfus describes how he was able to trick ELIZA, a computer program written by Joseph Weizenbaum to simulate a psychotherapist (see “ELIZA Passes the Turing Test,” below).60 Aside from having been written twenty years earlier, ELIZA was considered even then to be a simpleminded program unrepresentative of the state of the art.61 Perhaps Dreyfus’ most consistent theme is the inability of the hard antecedent-consequence type of logic to solve certain types of problems.62 While this observation is correct, most AI researchers do not propose “PROLOG-like” logic as the solution to all problems.63 For example, using fuzzy logic principles to deal in a methodologically sound manner with uncertain observations is becoming increasingly popular.64 It is also feasible to create systems with thousands or even millions of parallel processes to emulate human pattern recognition and skill acquisition abilities. Dreyfus describes machine intelligence as a fixed phenomenon and regards today’s apparent limitations as permanent limitations rather than tomorrow’s challenges.
Dreyfus continually presses the theme that the AI field has been overly optimistic and has underestimated the deep nature of many problems. The criticism has considerable justification, but Dreyfus takes the position to an extreme by listing specific tasks that he maintains a computer will never do, including playing championship chess, riding a bicycle, and understanding human speech.65 These would seem
Photo by Lou Jones www.fotojones.com
to be needlessly negative predictions, reminiscent of earlier predictions that “man is not meant to fly” (“man,” of course, does a lot of things he was “not meant” to do). Computers are close to accomplishing some of these tasks now, though how close depends, of course, on the standard of performance one will accept.66 In my opinion, the levels of machine performance will continue to increase over time and, if one is intellectually honest, it appears to be only a matter of time before any particular standard is reached. And perhaps if Dreyfus is intellectually honest, he will be won over. He has in fact stated that if in a Turing test, a machine could fool him 60 percent of the time as to whether he was dealing with natural or artificial intelligence, he would concede defeat.67 One has to note, however, that this offer is less generous than it appears at first glance. Dreyfus, as the human judge in such a Turing test, would be able to achieve an accuracy of 50 percent just by guessing randomly.
In turn, AI critics of Dreyfus have been quick to jump on Dreyfus’s limited understanding of computer technology but have themselves not taken adequate time to understand either Dreyfus’s intellectual tradition, with its roots in the work of Kierkegaard, Heidegger, and the late Wittgenstein, or the relevance of this tradition to the goals of the AI movement.68
In a recent article by Hubert Dreyfus and his brother Stuart Dreyfus, the reader detects a subtle but possibly significant shift in Dreyfus’s approach to machine intelligence.69 They begin with the following quotes: “Nothing seems more possible to me than that people some day will come to the definite opinion that there is no copy in the . . . nervous system which corresponds to a particular thought, or a particular idea, or memory” (Ludwig Wittgenstein 1948). “Information is not stored anywhere in particular. Rather, it is stored everywhere. Information is better thought of as evoked than found” (David Rumelhart and Donald Norman 1981).70
The article goes on to state their continued strong opposition to a concept of artificial intelligence based entirely on symbolic reasoning, but the authors appear to be more comfortable with broader notions of machine intelligence. In particular, the ability of a neural net (a special type of computer composed of many parallel processes, each of which simulates a human brain neuron) to produce unexpected associations appears to be intriguing to the Dreyfus brothers.71 They end by saying,
Perhaps a net must share size, architecture, and initial-connection configuration with the human brain if it is to share our sense of appropriate generalization. If it is to learn from its own “experiences” to make associations that are humanlike rather than be taught to make associations that have been specified by its trainer, a net must also share our sense of appropriateness of output, and this means it must share our needs, desires, and emotions and have a humanlike body with appropriate physical movements, abilities, and vulnerability to injury…. If the minimum unit of analysis is that of a whole organism geared into a whole cultural world, neural nets as well as symbolically programmed computers still have a very long way to go.72
While the Dreyfuses’ insistence that an intelligence must share a humanlike body to truly emulate human intelligence has always caused consternation among AI theorists, the above statement seems to be more accepting than previous writings of at least the theoretical possibility of endowing a machine with true intelligence .73
I felt it would be worthwhile to explore philosophical issues of machine cognition with a number of experts who, while having extensive computer experience, were not influenced by two thousand years of theory and debate.74 I therefore chose six children ages seven to nine who had been working with computers for several years, but who assured me that they had not read either the early or the late Wittgenstein.
I told each child (one at a time) that I would ask them several questions and that there was no right or wrong answer, I just wanted their opinion. The questions were, Can a computer remember? Does a computer learn? Do computers think? Do computers have feelings? Do you like computers? Do computers like you?
To the first two questions each child answered in the affirmative: computers do remember, and they do learn. The third question required a few moments of reflection, and all but two of the children concluded that yes, computers do think. Apparently, one important clue to computers’ thinking ability for the children was the fact that when the children ask a computer to do something, it sometimes answers right away and sometimes there is a delay while the computer apparently thinks about the task for a while before responding. This, the children felt, was very similar to the way that they respond to questions.
The fourth question-Do computers have feelings?- not only was unanimously answered in the negative; it generally elicited laughter as if I had asked, Do elephants fly? Laughter, according to Freud, sometimes results from the juxtaposition of two concepts that are not supposed to go together, which may result from either the two concepts’ never having ever been linked before or a social taboo.75 Possibly both reasons caused the laughter in this case.
On the fifth question, all of the children answered affirmatively that they liked computers. All of the children thought that the last question was silly, that computers do not have likes and dislikes.
Photo by Lou Jones www.fotojones.com